Fast and Robust Initialization for Visual-Inertial SLAM

Abstract

Visual-inertial SLAM (VI-SLAM) requires a good initial estimation of the initial velocity, orientation with respect to gravity and gyroscope and accelerometer biases. In this paper we build on the initialization method proposed by Martinelli [1] and extended by Kaiser et al. [2], modifying it to be more general and efficient. We improve accuracy with several rounds of visual-inertial bundle adjustment, and robustify the method with novel observability and consensus tests, that discard erroneous solutions. Our results on the EuRoC dataset show that, while the original method produces scale errors up to 156%, our method is able to consistently initialize in less than two seconds with scale errors around 5%, which can be further reduced to less than 1% performing visual-inertial bundle adjustment after ten seconds.

3D LiDAR/IMU Calibration Based on Continuous-Time Trajectory Estimation in Structured Environments

“…In general, the multimodal sensor calibration problem can be solved by modeling it as an optimization problem. The importance of the accuracy of optimization initials has been thoroughly discussed in many works [24]- [26] , which have demonstrated that an accurate optimization result requires a robust initialization procedure to obtain reliable initials. Consequently, the two-step calibration scheme has become the most popular framework for auto-calibration of multisensors.…”

Optimization-Based Online Initialization and Calibration of Monocular Visual-Inertial Odometry Considering Spatial-Temporal Constraints

“…To overcome the disadvantages, several online methods were developed. The online methods assumed that the measurements from a camera and an IMU were well synchronized (e.g., [14][15][16][17][18][19]), or the extrinsic spatial parameter was known in advance (e.g., [20][21][22]), or both conditions were satisfied (e.g., [23][24][25][26][27][28]). In the case where both the measurements from different sensors are asynchronous and the extrinsic spatial parameter between different sensors is unknown, most of the existing methods [29][30][31][32][33][34] are designed for filter-based VIOs since they are usually built on the Multi-State Constraint Kalman Filter (MSCKF [35]) framework.…”

Fast Monocular Visual‐Inertial Initialization with an Improved Iterative Strategy

“…In early studies, several initialization methods have been studied, such as the representative joint methods [17][18][19] and disjoint methods [16,20,21].…”

“…In later research, it is improved by Kaiser et al with a little bit of precision that is sacrificed [18]. The method in [19] suffers a low initialization recall; it only works in twenty percent of the trajectory points, which might be a problem for the robot applications in case the system needs to be launched immediately (ii) The disjoint method is first introduced by Murartal and Tardos [16] and latter adapted by Qin and Shen and Yang and Shen [20,21] with a good performance. In both cases, the parameters of IMU are estimated in different steps by solving a series of linear formulas with the least-squares method such as Gauss-Newton (G-N) and Levenberg-Marquardt (L-M) [22,23].…”